Output Image Size error

yashrunwal

I have trained a custom YOLOv5 model and then converted it to .blob using the online converter tool. The model is trained to detect only one class. Then I placed this model along with custom_model.json file. I changed the contents of the json file to include only one class and the input size: 640x640.

Then ran the depthai_demo.py script and selected 'custom_model' from the CNN dropdown menu. But then I am getting the following error:

What does this mean?

Thank You.

erik

Hello yashrunwal ,
sorry about the delay. You can set the ImageManip's maximum frame size like this: manip.setMaxOutputFrameSize(1228800) and it should work as expected. We will work on providing better error message so it's clear how to fix this.
Thanks, Erik

yashrunwal

erik Hello, As you suggested, I used this command manip.setMaxOutputFrameSize(1228800) in the pipeline_manager.py as shown below:

However, the error still exists. Any help?

Edit 1:
I tried to use the aforementioned line directly in the depthai_demo.py script and I got similar results:

erik

Hello yashrunwal , did it change the error message, so the second number is higher, or is it exactly the same?

yashrunwal

Hi erik It is exactly the same

Edit:
I used this script (https://github.com/luxonis/depthai-experiments/blob/master/gen2-yolo/device-decoding/main.py) and added the following code to it: pm.pipeline.create(dai.node.ImageManip).setMaxOutputFrameSize(1228800).
Then ran the code, but still get the same error.

erik

Hello yashrunwal , asking the team internally so we add support for larger frames to the depthai_demo as well, it should be added this week. Will also circle back with the temporary solution.
Thanks, Erik

Luxonis-Lukasz

Hi yashrunwal
Could you try the fix_nn_manip branch? I created a draft PR with a fix that should work for you. Let me know if this helps.

To use it, in the depthai repo root, run these commands:

git pull
git checkout fix_nn_manip

yashrunwal

Hey @Luxonis-Lukasz It worked. Thanks for your help!!